Google is working on a new server chip that would directly integrate components of its Gemini model into the hardware in an effort to provide its AI models to customers more effectively. The Information reported on Monday.
The new chip, unofficially known as “Frozen v2,” is expected by the Alphabet-owned corporation to help alleviate an AI computing capacity shortage that has exacerbated internal conflicts and led Google Cloud to reject contracts with external clients, according to the report.
Early trading saw a 3.3% increase in Alphabet shares.
Here are some specifics:
Google intends to release the chip as early as 2028, but engineers are still working on the architecture and the quantity of model data that will be hardwired.
Based on the quantity of AI tokens served per unit of electricity, the chip may be six to ten times more efficient than Google’s most recent proprietary AI processors.
Our teams are always investigating and testing novel ideas. A Google Cloud representative stated, “We make sure our systems are integrated and highly optimized by co-designing our hardware and software from the ground up.”
Instead of replacing Google’s tensor processing units (TPUs), the “Frozen” project aims to develop a new set of native chips, according to the article.
Last week, Bloomberg News revealed that Google postponed the release of its most recent Gemini AI model because it did not meet internal targets.
The business is currently attempting to enhance its capabilities, especially in coding.
